YoloCSharp - Dotnet wrapper for Yolo

YoloCSharp is a cross platform wrapper of Yolo/Darknet for .Net Standard 2.0.

• Warning: Only 64 bits;
• Support for Windows/Linux/Docker;
• Compatible with .Net Framework 4.6.1+/.Net Core 2.0;
• GPU Computation with CUDA;
• No CPU Computation available.

Features

• Easy to use;
• Use Yolo/Darknet library from the repo of AlexeyAB;
• Use OpenCV 3.3.1 to able to deal with the images or videos (See API);
• Compatible with Yolo v2 and Yolo v3.

Requirements

1. Install CUDA 10.0 and cuDNN 7.4.1;
2. Restart your computer;
3. On Linux, you need to have OpenCV 3.3.1. See your distribution documentation for more information.

Installation

Simply, add the OpenCvSharp-AnyCPU nuget in the version 3.3.1.20171117 and the YoloCSharp nuget` in your project.

Warning: You need to set your project in 64 bits. For Visual Studio 2017, see the docs.

Note: Due to a bug with OpenCV 3.4.1 and later, Yolo is not compatible with them. So these wrapper is only build for the version 3.3.1.

Examples

All examples are available in the examples folder. Don’t hesitate to create new and send them to us.

Note : You need to download yolov3.weights and put it in the data folder before test any example.

API

Darknet.cs

/// <summary>
/// Provides a Darknet detector which can be used to analyze image and find all detectable objects inside
/// </summary>
public class Darknet : IDisposable
{
	/// <summary>
	/// Initializes a new Darknet detector with the cfgFile and the weightFile
	/// </summary>
	/// <param name="cfgFile">The *.cfg file</param>
	/// <param name="weightFile">The *.weights file</param>
	/// <param name="gpuId">The index of a CUDA compatible GPU</param>
	public Darknet(string cfgFile, string weightFile, int gpuId = 0);

	/// <summary>
	/// Use to detect objects on an OpenCV Mat object and return all found objects
	/// </summary>
	/// <returns>List of YoloResult</returns>
	/// <param name="mat">The frame to analyze</param>
	/// <param name="thresh">Threshold at which an object should be confirmed</param>
	/// <param name="useMean">Unknown parameter</param>
	public List<YoloResult> Detect(Mat mat, float thresh=0.2f, bool useMean=false);
}

YoloResult.cs

/// <summary>
/// Represents the detection information of an object by Yolo
/// </summary>
public class YoloResult
{
	/// <summary>
	/// X coordinate of the top-left corner of bounded box
	/// </summary>
	public uint X { get; }

	/// <summary>
	/// Y coordinate of the top-left corner of bounded box
	/// </summary>
	public uint Y { get; }

	/// <summary>
	/// Width of bounded box
	/// </summary>
	public uint Width { get; }

	/// <summary>
	/// Height of bounded box
	/// </summary>
	public uint Height { get; }

	/// <summary>
	/// Probability that the object was found correctly
	/// </summary>
	public float Prob { get; }

	/// <summary>
	/// Class of object - from range [0, classes-1]
	/// </summary>
	public uint ObjId { get; }

	/// <summary>
	/// Tracking id for video (0 - untracked, 1 - inf - tracked object)
	/// </summary>
	public uint TrackId { get; }

	/// <summary>
	/// Counter of frames on which the object was detected
	/// </summary>
	public uint FramesCounter { get; }
}